tests: Fix a deeper bug in null byte check in testsuite.

Thanks to Jeremy Lavergne for reporting the problem.

diff --git a/tests/language/lexer/lexer.at b/tests/language/lexer/lexer.at
index 6d3a5ad9860b0581b0305b9ba8a41fa0d5f48983..bd960710c47686a3ade9a2a02715dbef06829691 100644 (file)
--- a/tests/language/lexer/lexer.at
+++ b/tests/language/lexer/lexer.at
@@ -71,10 +71,13 @@ lis|.\0" > lexer.sps
 # We sort the output into a predictable order because the lexer finds
 # and reports null bytes as soon as it reads them into its input
 # buffer, as opposed to when it encounters them during tokenization.
+# This also means that null bytes might be reported as part of one
+# command or another or none, hence removing the LIST: prefix.
 AT_CHECK([pspp -O format=csv lexer.sps > lexer.csv], [1])
-AT_CHECK([sed '/^$/d' lexer.csv | sort], [0], [dnl
+AT_CHECK([sed '/^$/d
+s/LIST: //' lexer.csv | sort], [0], [dnl
 lexer.sps: error: Bad character U+0000 in input.
 lexer.sps:1: error: Unknown command `datA dist'.
-lexer.sps:2: error: LIST: LIST is allowed only after the active dataset has been defined.
+lexer.sps:2: error: LIST is allowed only after the active dataset has been defined.
 ])
 AT_CLEANUP